Safe Online Control-Informed Learning

2025-12-15

AI Summary

To address the challenge of online adaptive control for safety-critical autonomous systems operating in uncertain environments, this paper proposes a safety-constrained online learning framework. Methodologically, it integrates optimal control, parameter adaptive estimation, and extended Kalman filtering, and innovatively introduces a softplus barrier function to embed safety constraints in an initial-condition-independent manner—rigorously proving both convergence and formal safety guarantees. The framework enables efficient, robust, control-guided online learning. Experimental validation on inverted pendulum and robotic arm tasks demonstrates significant improvements: approximately 40% higher data efficiency and a safety constraint satisfaction rate of 99.2%, substantially outperforming baseline approaches. Crucially, it eliminates the strong reliance on an initially safe policy inherent in conventional methods. This work establishes a verifiable, deployable paradigm for safety-driven autonomous learning.

Technology Category

• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Online Learning & Bandits

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

This paper proposes a Safe Online Control-Informed Learning framework for safety-critical autonomous systems. The framework unifies optimal control, parameter estimation, and safety constraints into an online learning process. It employs an extended Kalman filter to incrementally update system parameters in real time, enabling robust and data-efficient adaptation under uncertainty. A softplus barrier function enforces constraint satisfaction during learning and control while eliminating the dependence on high-quality initial guesses. Theoretical analysis establishes convergence and safety guarantees, and the framework's effectiveness is demonstrated on cart-pole and robot-arm systems.

Problem

• Ensures safety in online learning for autonomous systems
• Unifies control, estimation, and constraints for real-time adaptation
• Guarantees convergence and safety with theoretical and experimental validation

Innovation

• Online learning unifies control, estimation, and safety constraints
• Extended Kalman filter updates parameters incrementally in real time
• Softplus barrier function enforces constraints without initial guess dependence